xref: /openbmc/linux/tools/perf/builtin-sched.c (revision bc000245)
1 #include "builtin.h"
2 #include "perf.h"
3 
4 #include "util/util.h"
5 #include "util/evlist.h"
6 #include "util/cache.h"
7 #include "util/evsel.h"
8 #include "util/symbol.h"
9 #include "util/thread.h"
10 #include "util/header.h"
11 #include "util/session.h"
12 #include "util/tool.h"
13 
14 #include "util/parse-options.h"
15 #include "util/trace-event.h"
16 
17 #include "util/debug.h"
18 
19 #include <sys/prctl.h>
20 #include <sys/resource.h>
21 
22 #include <semaphore.h>
23 #include <pthread.h>
24 #include <math.h>
25 
26 #define PR_SET_NAME		15               /* Set process name */
27 #define MAX_CPUS		4096
28 #define COMM_LEN		20
29 #define SYM_LEN			129
30 #define MAX_PID			65536
31 
32 struct sched_atom;
33 
34 struct task_desc {
35 	unsigned long		nr;
36 	unsigned long		pid;
37 	char			comm[COMM_LEN];
38 
39 	unsigned long		nr_events;
40 	unsigned long		curr_event;
41 	struct sched_atom	**atoms;
42 
43 	pthread_t		thread;
44 	sem_t			sleep_sem;
45 
46 	sem_t			ready_for_work;
47 	sem_t			work_done_sem;
48 
49 	u64			cpu_usage;
50 };
51 
52 enum sched_event_type {
53 	SCHED_EVENT_RUN,
54 	SCHED_EVENT_SLEEP,
55 	SCHED_EVENT_WAKEUP,
56 	SCHED_EVENT_MIGRATION,
57 };
58 
59 struct sched_atom {
60 	enum sched_event_type	type;
61 	int			specific_wait;
62 	u64			timestamp;
63 	u64			duration;
64 	unsigned long		nr;
65 	sem_t			*wait_sem;
66 	struct task_desc	*wakee;
67 };
68 
69 #define TASK_STATE_TO_CHAR_STR "RSDTtZX"
70 
71 enum thread_state {
72 	THREAD_SLEEPING = 0,
73 	THREAD_WAIT_CPU,
74 	THREAD_SCHED_IN,
75 	THREAD_IGNORE
76 };
77 
78 struct work_atom {
79 	struct list_head	list;
80 	enum thread_state	state;
81 	u64			sched_out_time;
82 	u64			wake_up_time;
83 	u64			sched_in_time;
84 	u64			runtime;
85 };
86 
87 struct work_atoms {
88 	struct list_head	work_list;
89 	struct thread		*thread;
90 	struct rb_node		node;
91 	u64			max_lat;
92 	u64			max_lat_at;
93 	u64			total_lat;
94 	u64			nb_atoms;
95 	u64			total_runtime;
96 };
97 
98 typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *);
99 
100 struct perf_sched;
101 
102 struct trace_sched_handler {
103 	int (*switch_event)(struct perf_sched *sched, struct perf_evsel *evsel,
104 			    struct perf_sample *sample, struct machine *machine);
105 
106 	int (*runtime_event)(struct perf_sched *sched, struct perf_evsel *evsel,
107 			     struct perf_sample *sample, struct machine *machine);
108 
109 	int (*wakeup_event)(struct perf_sched *sched, struct perf_evsel *evsel,
110 			    struct perf_sample *sample, struct machine *machine);
111 
112 	/* PERF_RECORD_FORK event, not sched_process_fork tracepoint */
113 	int (*fork_event)(struct perf_sched *sched, union perf_event *event,
114 			  struct machine *machine);
115 
116 	int (*migrate_task_event)(struct perf_sched *sched,
117 				  struct perf_evsel *evsel,
118 				  struct perf_sample *sample,
119 				  struct machine *machine);
120 };
121 
122 struct perf_sched {
123 	struct perf_tool tool;
124 	const char	 *sort_order;
125 	unsigned long	 nr_tasks;
126 	struct task_desc *pid_to_task[MAX_PID];
127 	struct task_desc **tasks;
128 	const struct trace_sched_handler *tp_handler;
129 	pthread_mutex_t	 start_work_mutex;
130 	pthread_mutex_t	 work_done_wait_mutex;
131 	int		 profile_cpu;
132 /*
133  * Track the current task - that way we can know whether there's any
134  * weird events, such as a task being switched away that is not current.
135  */
136 	int		 max_cpu;
137 	u32		 curr_pid[MAX_CPUS];
138 	struct thread	 *curr_thread[MAX_CPUS];
139 	char		 next_shortname1;
140 	char		 next_shortname2;
141 	unsigned int	 replay_repeat;
142 	unsigned long	 nr_run_events;
143 	unsigned long	 nr_sleep_events;
144 	unsigned long	 nr_wakeup_events;
145 	unsigned long	 nr_sleep_corrections;
146 	unsigned long	 nr_run_events_optimized;
147 	unsigned long	 targetless_wakeups;
148 	unsigned long	 multitarget_wakeups;
149 	unsigned long	 nr_runs;
150 	unsigned long	 nr_timestamps;
151 	unsigned long	 nr_unordered_timestamps;
152 	unsigned long	 nr_state_machine_bugs;
153 	unsigned long	 nr_context_switch_bugs;
154 	unsigned long	 nr_events;
155 	unsigned long	 nr_lost_chunks;
156 	unsigned long	 nr_lost_events;
157 	u64		 run_measurement_overhead;
158 	u64		 sleep_measurement_overhead;
159 	u64		 start_time;
160 	u64		 cpu_usage;
161 	u64		 runavg_cpu_usage;
162 	u64		 parent_cpu_usage;
163 	u64		 runavg_parent_cpu_usage;
164 	u64		 sum_runtime;
165 	u64		 sum_fluct;
166 	u64		 run_avg;
167 	u64		 all_runtime;
168 	u64		 all_count;
169 	u64		 cpu_last_switched[MAX_CPUS];
170 	struct rb_root	 atom_root, sorted_atom_root;
171 	struct list_head sort_list, cmp_pid;
172 };
173 
174 static u64 get_nsecs(void)
175 {
176 	struct timespec ts;
177 
178 	clock_gettime(CLOCK_MONOTONIC, &ts);
179 
180 	return ts.tv_sec * 1000000000ULL + ts.tv_nsec;
181 }
182 
183 static void burn_nsecs(struct perf_sched *sched, u64 nsecs)
184 {
185 	u64 T0 = get_nsecs(), T1;
186 
187 	do {
188 		T1 = get_nsecs();
189 	} while (T1 + sched->run_measurement_overhead < T0 + nsecs);
190 }
191 
192 static void sleep_nsecs(u64 nsecs)
193 {
194 	struct timespec ts;
195 
196 	ts.tv_nsec = nsecs % 999999999;
197 	ts.tv_sec = nsecs / 999999999;
198 
199 	nanosleep(&ts, NULL);
200 }
201 
202 static void calibrate_run_measurement_overhead(struct perf_sched *sched)
203 {
204 	u64 T0, T1, delta, min_delta = 1000000000ULL;
205 	int i;
206 
207 	for (i = 0; i < 10; i++) {
208 		T0 = get_nsecs();
209 		burn_nsecs(sched, 0);
210 		T1 = get_nsecs();
211 		delta = T1-T0;
212 		min_delta = min(min_delta, delta);
213 	}
214 	sched->run_measurement_overhead = min_delta;
215 
216 	printf("run measurement overhead: %" PRIu64 " nsecs\n", min_delta);
217 }
218 
219 static void calibrate_sleep_measurement_overhead(struct perf_sched *sched)
220 {
221 	u64 T0, T1, delta, min_delta = 1000000000ULL;
222 	int i;
223 
224 	for (i = 0; i < 10; i++) {
225 		T0 = get_nsecs();
226 		sleep_nsecs(10000);
227 		T1 = get_nsecs();
228 		delta = T1-T0;
229 		min_delta = min(min_delta, delta);
230 	}
231 	min_delta -= 10000;
232 	sched->sleep_measurement_overhead = min_delta;
233 
234 	printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta);
235 }
236 
237 static struct sched_atom *
238 get_new_event(struct task_desc *task, u64 timestamp)
239 {
240 	struct sched_atom *event = zalloc(sizeof(*event));
241 	unsigned long idx = task->nr_events;
242 	size_t size;
243 
244 	event->timestamp = timestamp;
245 	event->nr = idx;
246 
247 	task->nr_events++;
248 	size = sizeof(struct sched_atom *) * task->nr_events;
249 	task->atoms = realloc(task->atoms, size);
250 	BUG_ON(!task->atoms);
251 
252 	task->atoms[idx] = event;
253 
254 	return event;
255 }
256 
257 static struct sched_atom *last_event(struct task_desc *task)
258 {
259 	if (!task->nr_events)
260 		return NULL;
261 
262 	return task->atoms[task->nr_events - 1];
263 }
264 
265 static void add_sched_event_run(struct perf_sched *sched, struct task_desc *task,
266 				u64 timestamp, u64 duration)
267 {
268 	struct sched_atom *event, *curr_event = last_event(task);
269 
270 	/*
271 	 * optimize an existing RUN event by merging this one
272 	 * to it:
273 	 */
274 	if (curr_event && curr_event->type == SCHED_EVENT_RUN) {
275 		sched->nr_run_events_optimized++;
276 		curr_event->duration += duration;
277 		return;
278 	}
279 
280 	event = get_new_event(task, timestamp);
281 
282 	event->type = SCHED_EVENT_RUN;
283 	event->duration = duration;
284 
285 	sched->nr_run_events++;
286 }
287 
288 static void add_sched_event_wakeup(struct perf_sched *sched, struct task_desc *task,
289 				   u64 timestamp, struct task_desc *wakee)
290 {
291 	struct sched_atom *event, *wakee_event;
292 
293 	event = get_new_event(task, timestamp);
294 	event->type = SCHED_EVENT_WAKEUP;
295 	event->wakee = wakee;
296 
297 	wakee_event = last_event(wakee);
298 	if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) {
299 		sched->targetless_wakeups++;
300 		return;
301 	}
302 	if (wakee_event->wait_sem) {
303 		sched->multitarget_wakeups++;
304 		return;
305 	}
306 
307 	wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem));
308 	sem_init(wakee_event->wait_sem, 0, 0);
309 	wakee_event->specific_wait = 1;
310 	event->wait_sem = wakee_event->wait_sem;
311 
312 	sched->nr_wakeup_events++;
313 }
314 
315 static void add_sched_event_sleep(struct perf_sched *sched, struct task_desc *task,
316 				  u64 timestamp, u64 task_state __maybe_unused)
317 {
318 	struct sched_atom *event = get_new_event(task, timestamp);
319 
320 	event->type = SCHED_EVENT_SLEEP;
321 
322 	sched->nr_sleep_events++;
323 }
324 
325 static struct task_desc *register_pid(struct perf_sched *sched,
326 				      unsigned long pid, const char *comm)
327 {
328 	struct task_desc *task;
329 
330 	BUG_ON(pid >= MAX_PID);
331 
332 	task = sched->pid_to_task[pid];
333 
334 	if (task)
335 		return task;
336 
337 	task = zalloc(sizeof(*task));
338 	task->pid = pid;
339 	task->nr = sched->nr_tasks;
340 	strcpy(task->comm, comm);
341 	/*
342 	 * every task starts in sleeping state - this gets ignored
343 	 * if there's no wakeup pointing to this sleep state:
344 	 */
345 	add_sched_event_sleep(sched, task, 0, 0);
346 
347 	sched->pid_to_task[pid] = task;
348 	sched->nr_tasks++;
349 	sched->tasks = realloc(sched->tasks, sched->nr_tasks * sizeof(struct task_task *));
350 	BUG_ON(!sched->tasks);
351 	sched->tasks[task->nr] = task;
352 
353 	if (verbose)
354 		printf("registered task #%ld, PID %ld (%s)\n", sched->nr_tasks, pid, comm);
355 
356 	return task;
357 }
358 
359 
360 static void print_task_traces(struct perf_sched *sched)
361 {
362 	struct task_desc *task;
363 	unsigned long i;
364 
365 	for (i = 0; i < sched->nr_tasks; i++) {
366 		task = sched->tasks[i];
367 		printf("task %6ld (%20s:%10ld), nr_events: %ld\n",
368 			task->nr, task->comm, task->pid, task->nr_events);
369 	}
370 }
371 
372 static void add_cross_task_wakeups(struct perf_sched *sched)
373 {
374 	struct task_desc *task1, *task2;
375 	unsigned long i, j;
376 
377 	for (i = 0; i < sched->nr_tasks; i++) {
378 		task1 = sched->tasks[i];
379 		j = i + 1;
380 		if (j == sched->nr_tasks)
381 			j = 0;
382 		task2 = sched->tasks[j];
383 		add_sched_event_wakeup(sched, task1, 0, task2);
384 	}
385 }
386 
387 static void perf_sched__process_event(struct perf_sched *sched,
388 				      struct sched_atom *atom)
389 {
390 	int ret = 0;
391 
392 	switch (atom->type) {
393 		case SCHED_EVENT_RUN:
394 			burn_nsecs(sched, atom->duration);
395 			break;
396 		case SCHED_EVENT_SLEEP:
397 			if (atom->wait_sem)
398 				ret = sem_wait(atom->wait_sem);
399 			BUG_ON(ret);
400 			break;
401 		case SCHED_EVENT_WAKEUP:
402 			if (atom->wait_sem)
403 				ret = sem_post(atom->wait_sem);
404 			BUG_ON(ret);
405 			break;
406 		case SCHED_EVENT_MIGRATION:
407 			break;
408 		default:
409 			BUG_ON(1);
410 	}
411 }
412 
413 static u64 get_cpu_usage_nsec_parent(void)
414 {
415 	struct rusage ru;
416 	u64 sum;
417 	int err;
418 
419 	err = getrusage(RUSAGE_SELF, &ru);
420 	BUG_ON(err);
421 
422 	sum =  ru.ru_utime.tv_sec*1e9 + ru.ru_utime.tv_usec*1e3;
423 	sum += ru.ru_stime.tv_sec*1e9 + ru.ru_stime.tv_usec*1e3;
424 
425 	return sum;
426 }
427 
428 static int self_open_counters(void)
429 {
430 	struct perf_event_attr attr;
431 	int fd;
432 
433 	memset(&attr, 0, sizeof(attr));
434 
435 	attr.type = PERF_TYPE_SOFTWARE;
436 	attr.config = PERF_COUNT_SW_TASK_CLOCK;
437 
438 	fd = sys_perf_event_open(&attr, 0, -1, -1, 0);
439 
440 	if (fd < 0)
441 		pr_err("Error: sys_perf_event_open() syscall returned "
442 		       "with %d (%s)\n", fd, strerror(errno));
443 	return fd;
444 }
445 
446 static u64 get_cpu_usage_nsec_self(int fd)
447 {
448 	u64 runtime;
449 	int ret;
450 
451 	ret = read(fd, &runtime, sizeof(runtime));
452 	BUG_ON(ret != sizeof(runtime));
453 
454 	return runtime;
455 }
456 
457 struct sched_thread_parms {
458 	struct task_desc  *task;
459 	struct perf_sched *sched;
460 };
461 
462 static void *thread_func(void *ctx)
463 {
464 	struct sched_thread_parms *parms = ctx;
465 	struct task_desc *this_task = parms->task;
466 	struct perf_sched *sched = parms->sched;
467 	u64 cpu_usage_0, cpu_usage_1;
468 	unsigned long i, ret;
469 	char comm2[22];
470 	int fd;
471 
472 	free(parms);
473 
474 	sprintf(comm2, ":%s", this_task->comm);
475 	prctl(PR_SET_NAME, comm2);
476 	fd = self_open_counters();
477 	if (fd < 0)
478 		return NULL;
479 again:
480 	ret = sem_post(&this_task->ready_for_work);
481 	BUG_ON(ret);
482 	ret = pthread_mutex_lock(&sched->start_work_mutex);
483 	BUG_ON(ret);
484 	ret = pthread_mutex_unlock(&sched->start_work_mutex);
485 	BUG_ON(ret);
486 
487 	cpu_usage_0 = get_cpu_usage_nsec_self(fd);
488 
489 	for (i = 0; i < this_task->nr_events; i++) {
490 		this_task->curr_event = i;
491 		perf_sched__process_event(sched, this_task->atoms[i]);
492 	}
493 
494 	cpu_usage_1 = get_cpu_usage_nsec_self(fd);
495 	this_task->cpu_usage = cpu_usage_1 - cpu_usage_0;
496 	ret = sem_post(&this_task->work_done_sem);
497 	BUG_ON(ret);
498 
499 	ret = pthread_mutex_lock(&sched->work_done_wait_mutex);
500 	BUG_ON(ret);
501 	ret = pthread_mutex_unlock(&sched->work_done_wait_mutex);
502 	BUG_ON(ret);
503 
504 	goto again;
505 }
506 
507 static void create_tasks(struct perf_sched *sched)
508 {
509 	struct task_desc *task;
510 	pthread_attr_t attr;
511 	unsigned long i;
512 	int err;
513 
514 	err = pthread_attr_init(&attr);
515 	BUG_ON(err);
516 	err = pthread_attr_setstacksize(&attr,
517 			(size_t) max(16 * 1024, PTHREAD_STACK_MIN));
518 	BUG_ON(err);
519 	err = pthread_mutex_lock(&sched->start_work_mutex);
520 	BUG_ON(err);
521 	err = pthread_mutex_lock(&sched->work_done_wait_mutex);
522 	BUG_ON(err);
523 	for (i = 0; i < sched->nr_tasks; i++) {
524 		struct sched_thread_parms *parms = malloc(sizeof(*parms));
525 		BUG_ON(parms == NULL);
526 		parms->task = task = sched->tasks[i];
527 		parms->sched = sched;
528 		sem_init(&task->sleep_sem, 0, 0);
529 		sem_init(&task->ready_for_work, 0, 0);
530 		sem_init(&task->work_done_sem, 0, 0);
531 		task->curr_event = 0;
532 		err = pthread_create(&task->thread, &attr, thread_func, parms);
533 		BUG_ON(err);
534 	}
535 }
536 
537 static void wait_for_tasks(struct perf_sched *sched)
538 {
539 	u64 cpu_usage_0, cpu_usage_1;
540 	struct task_desc *task;
541 	unsigned long i, ret;
542 
543 	sched->start_time = get_nsecs();
544 	sched->cpu_usage = 0;
545 	pthread_mutex_unlock(&sched->work_done_wait_mutex);
546 
547 	for (i = 0; i < sched->nr_tasks; i++) {
548 		task = sched->tasks[i];
549 		ret = sem_wait(&task->ready_for_work);
550 		BUG_ON(ret);
551 		sem_init(&task->ready_for_work, 0, 0);
552 	}
553 	ret = pthread_mutex_lock(&sched->work_done_wait_mutex);
554 	BUG_ON(ret);
555 
556 	cpu_usage_0 = get_cpu_usage_nsec_parent();
557 
558 	pthread_mutex_unlock(&sched->start_work_mutex);
559 
560 	for (i = 0; i < sched->nr_tasks; i++) {
561 		task = sched->tasks[i];
562 		ret = sem_wait(&task->work_done_sem);
563 		BUG_ON(ret);
564 		sem_init(&task->work_done_sem, 0, 0);
565 		sched->cpu_usage += task->cpu_usage;
566 		task->cpu_usage = 0;
567 	}
568 
569 	cpu_usage_1 = get_cpu_usage_nsec_parent();
570 	if (!sched->runavg_cpu_usage)
571 		sched->runavg_cpu_usage = sched->cpu_usage;
572 	sched->runavg_cpu_usage = (sched->runavg_cpu_usage * 9 + sched->cpu_usage) / 10;
573 
574 	sched->parent_cpu_usage = cpu_usage_1 - cpu_usage_0;
575 	if (!sched->runavg_parent_cpu_usage)
576 		sched->runavg_parent_cpu_usage = sched->parent_cpu_usage;
577 	sched->runavg_parent_cpu_usage = (sched->runavg_parent_cpu_usage * 9 +
578 					 sched->parent_cpu_usage)/10;
579 
580 	ret = pthread_mutex_lock(&sched->start_work_mutex);
581 	BUG_ON(ret);
582 
583 	for (i = 0; i < sched->nr_tasks; i++) {
584 		task = sched->tasks[i];
585 		sem_init(&task->sleep_sem, 0, 0);
586 		task->curr_event = 0;
587 	}
588 }
589 
590 static void run_one_test(struct perf_sched *sched)
591 {
592 	u64 T0, T1, delta, avg_delta, fluct;
593 
594 	T0 = get_nsecs();
595 	wait_for_tasks(sched);
596 	T1 = get_nsecs();
597 
598 	delta = T1 - T0;
599 	sched->sum_runtime += delta;
600 	sched->nr_runs++;
601 
602 	avg_delta = sched->sum_runtime / sched->nr_runs;
603 	if (delta < avg_delta)
604 		fluct = avg_delta - delta;
605 	else
606 		fluct = delta - avg_delta;
607 	sched->sum_fluct += fluct;
608 	if (!sched->run_avg)
609 		sched->run_avg = delta;
610 	sched->run_avg = (sched->run_avg * 9 + delta) / 10;
611 
612 	printf("#%-3ld: %0.3f, ", sched->nr_runs, (double)delta / 1000000.0);
613 
614 	printf("ravg: %0.2f, ", (double)sched->run_avg / 1e6);
615 
616 	printf("cpu: %0.2f / %0.2f",
617 		(double)sched->cpu_usage / 1e6, (double)sched->runavg_cpu_usage / 1e6);
618 
619 #if 0
620 	/*
621 	 * rusage statistics done by the parent, these are less
622 	 * accurate than the sched->sum_exec_runtime based statistics:
623 	 */
624 	printf(" [%0.2f / %0.2f]",
625 		(double)sched->parent_cpu_usage/1e6,
626 		(double)sched->runavg_parent_cpu_usage/1e6);
627 #endif
628 
629 	printf("\n");
630 
631 	if (sched->nr_sleep_corrections)
632 		printf(" (%ld sleep corrections)\n", sched->nr_sleep_corrections);
633 	sched->nr_sleep_corrections = 0;
634 }
635 
636 static void test_calibrations(struct perf_sched *sched)
637 {
638 	u64 T0, T1;
639 
640 	T0 = get_nsecs();
641 	burn_nsecs(sched, 1e6);
642 	T1 = get_nsecs();
643 
644 	printf("the run test took %" PRIu64 " nsecs\n", T1 - T0);
645 
646 	T0 = get_nsecs();
647 	sleep_nsecs(1e6);
648 	T1 = get_nsecs();
649 
650 	printf("the sleep test took %" PRIu64 " nsecs\n", T1 - T0);
651 }
652 
653 static int
654 replay_wakeup_event(struct perf_sched *sched,
655 		    struct perf_evsel *evsel, struct perf_sample *sample,
656 		    struct machine *machine __maybe_unused)
657 {
658 	const char *comm = perf_evsel__strval(evsel, sample, "comm");
659 	const u32 pid	 = perf_evsel__intval(evsel, sample, "pid");
660 	struct task_desc *waker, *wakee;
661 
662 	if (verbose) {
663 		printf("sched_wakeup event %p\n", evsel);
664 
665 		printf(" ... pid %d woke up %s/%d\n", sample->tid, comm, pid);
666 	}
667 
668 	waker = register_pid(sched, sample->tid, "<unknown>");
669 	wakee = register_pid(sched, pid, comm);
670 
671 	add_sched_event_wakeup(sched, waker, sample->time, wakee);
672 	return 0;
673 }
674 
675 static int replay_switch_event(struct perf_sched *sched,
676 			       struct perf_evsel *evsel,
677 			       struct perf_sample *sample,
678 			       struct machine *machine __maybe_unused)
679 {
680 	const char *prev_comm  = perf_evsel__strval(evsel, sample, "prev_comm"),
681 		   *next_comm  = perf_evsel__strval(evsel, sample, "next_comm");
682 	const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
683 		  next_pid = perf_evsel__intval(evsel, sample, "next_pid");
684 	const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
685 	struct task_desc *prev, __maybe_unused *next;
686 	u64 timestamp0, timestamp = sample->time;
687 	int cpu = sample->cpu;
688 	s64 delta;
689 
690 	if (verbose)
691 		printf("sched_switch event %p\n", evsel);
692 
693 	if (cpu >= MAX_CPUS || cpu < 0)
694 		return 0;
695 
696 	timestamp0 = sched->cpu_last_switched[cpu];
697 	if (timestamp0)
698 		delta = timestamp - timestamp0;
699 	else
700 		delta = 0;
701 
702 	if (delta < 0) {
703 		pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
704 		return -1;
705 	}
706 
707 	pr_debug(" ... switch from %s/%d to %s/%d [ran %" PRIu64 " nsecs]\n",
708 		 prev_comm, prev_pid, next_comm, next_pid, delta);
709 
710 	prev = register_pid(sched, prev_pid, prev_comm);
711 	next = register_pid(sched, next_pid, next_comm);
712 
713 	sched->cpu_last_switched[cpu] = timestamp;
714 
715 	add_sched_event_run(sched, prev, timestamp, delta);
716 	add_sched_event_sleep(sched, prev, timestamp, prev_state);
717 
718 	return 0;
719 }
720 
721 static int replay_fork_event(struct perf_sched *sched,
722 			     union perf_event *event,
723 			     struct machine *machine)
724 {
725 	struct thread *child, *parent;
726 
727 	child = machine__findnew_thread(machine, event->fork.pid,
728 					event->fork.tid);
729 	parent = machine__findnew_thread(machine, event->fork.ppid,
730 					 event->fork.ptid);
731 
732 	if (child == NULL || parent == NULL) {
733 		pr_debug("thread does not exist on fork event: child %p, parent %p\n",
734 				 child, parent);
735 		return 0;
736 	}
737 
738 	if (verbose) {
739 		printf("fork event\n");
740 		printf("... parent: %s/%d\n", thread__comm_str(parent), parent->tid);
741 		printf("...  child: %s/%d\n", thread__comm_str(child), child->tid);
742 	}
743 
744 	register_pid(sched, parent->tid, thread__comm_str(parent));
745 	register_pid(sched, child->tid, thread__comm_str(child));
746 	return 0;
747 }
748 
749 struct sort_dimension {
750 	const char		*name;
751 	sort_fn_t		cmp;
752 	struct list_head	list;
753 };
754 
755 static int
756 thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r)
757 {
758 	struct sort_dimension *sort;
759 	int ret = 0;
760 
761 	BUG_ON(list_empty(list));
762 
763 	list_for_each_entry(sort, list, list) {
764 		ret = sort->cmp(l, r);
765 		if (ret)
766 			return ret;
767 	}
768 
769 	return ret;
770 }
771 
772 static struct work_atoms *
773 thread_atoms_search(struct rb_root *root, struct thread *thread,
774 			 struct list_head *sort_list)
775 {
776 	struct rb_node *node = root->rb_node;
777 	struct work_atoms key = { .thread = thread };
778 
779 	while (node) {
780 		struct work_atoms *atoms;
781 		int cmp;
782 
783 		atoms = container_of(node, struct work_atoms, node);
784 
785 		cmp = thread_lat_cmp(sort_list, &key, atoms);
786 		if (cmp > 0)
787 			node = node->rb_left;
788 		else if (cmp < 0)
789 			node = node->rb_right;
790 		else {
791 			BUG_ON(thread != atoms->thread);
792 			return atoms;
793 		}
794 	}
795 	return NULL;
796 }
797 
798 static void
799 __thread_latency_insert(struct rb_root *root, struct work_atoms *data,
800 			 struct list_head *sort_list)
801 {
802 	struct rb_node **new = &(root->rb_node), *parent = NULL;
803 
804 	while (*new) {
805 		struct work_atoms *this;
806 		int cmp;
807 
808 		this = container_of(*new, struct work_atoms, node);
809 		parent = *new;
810 
811 		cmp = thread_lat_cmp(sort_list, data, this);
812 
813 		if (cmp > 0)
814 			new = &((*new)->rb_left);
815 		else
816 			new = &((*new)->rb_right);
817 	}
818 
819 	rb_link_node(&data->node, parent, new);
820 	rb_insert_color(&data->node, root);
821 }
822 
823 static int thread_atoms_insert(struct perf_sched *sched, struct thread *thread)
824 {
825 	struct work_atoms *atoms = zalloc(sizeof(*atoms));
826 	if (!atoms) {
827 		pr_err("No memory at %s\n", __func__);
828 		return -1;
829 	}
830 
831 	atoms->thread = thread;
832 	INIT_LIST_HEAD(&atoms->work_list);
833 	__thread_latency_insert(&sched->atom_root, atoms, &sched->cmp_pid);
834 	return 0;
835 }
836 
837 static char sched_out_state(u64 prev_state)
838 {
839 	const char *str = TASK_STATE_TO_CHAR_STR;
840 
841 	return str[prev_state];
842 }
843 
844 static int
845 add_sched_out_event(struct work_atoms *atoms,
846 		    char run_state,
847 		    u64 timestamp)
848 {
849 	struct work_atom *atom = zalloc(sizeof(*atom));
850 	if (!atom) {
851 		pr_err("Non memory at %s", __func__);
852 		return -1;
853 	}
854 
855 	atom->sched_out_time = timestamp;
856 
857 	if (run_state == 'R') {
858 		atom->state = THREAD_WAIT_CPU;
859 		atom->wake_up_time = atom->sched_out_time;
860 	}
861 
862 	list_add_tail(&atom->list, &atoms->work_list);
863 	return 0;
864 }
865 
866 static void
867 add_runtime_event(struct work_atoms *atoms, u64 delta,
868 		  u64 timestamp __maybe_unused)
869 {
870 	struct work_atom *atom;
871 
872 	BUG_ON(list_empty(&atoms->work_list));
873 
874 	atom = list_entry(atoms->work_list.prev, struct work_atom, list);
875 
876 	atom->runtime += delta;
877 	atoms->total_runtime += delta;
878 }
879 
880 static void
881 add_sched_in_event(struct work_atoms *atoms, u64 timestamp)
882 {
883 	struct work_atom *atom;
884 	u64 delta;
885 
886 	if (list_empty(&atoms->work_list))
887 		return;
888 
889 	atom = list_entry(atoms->work_list.prev, struct work_atom, list);
890 
891 	if (atom->state != THREAD_WAIT_CPU)
892 		return;
893 
894 	if (timestamp < atom->wake_up_time) {
895 		atom->state = THREAD_IGNORE;
896 		return;
897 	}
898 
899 	atom->state = THREAD_SCHED_IN;
900 	atom->sched_in_time = timestamp;
901 
902 	delta = atom->sched_in_time - atom->wake_up_time;
903 	atoms->total_lat += delta;
904 	if (delta > atoms->max_lat) {
905 		atoms->max_lat = delta;
906 		atoms->max_lat_at = timestamp;
907 	}
908 	atoms->nb_atoms++;
909 }
910 
911 static int latency_switch_event(struct perf_sched *sched,
912 				struct perf_evsel *evsel,
913 				struct perf_sample *sample,
914 				struct machine *machine)
915 {
916 	const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
917 		  next_pid = perf_evsel__intval(evsel, sample, "next_pid");
918 	const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
919 	struct work_atoms *out_events, *in_events;
920 	struct thread *sched_out, *sched_in;
921 	u64 timestamp0, timestamp = sample->time;
922 	int cpu = sample->cpu;
923 	s64 delta;
924 
925 	BUG_ON(cpu >= MAX_CPUS || cpu < 0);
926 
927 	timestamp0 = sched->cpu_last_switched[cpu];
928 	sched->cpu_last_switched[cpu] = timestamp;
929 	if (timestamp0)
930 		delta = timestamp - timestamp0;
931 	else
932 		delta = 0;
933 
934 	if (delta < 0) {
935 		pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
936 		return -1;
937 	}
938 
939 	sched_out = machine__findnew_thread(machine, 0, prev_pid);
940 	sched_in = machine__findnew_thread(machine, 0, next_pid);
941 
942 	out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid);
943 	if (!out_events) {
944 		if (thread_atoms_insert(sched, sched_out))
945 			return -1;
946 		out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid);
947 		if (!out_events) {
948 			pr_err("out-event: Internal tree error");
949 			return -1;
950 		}
951 	}
952 	if (add_sched_out_event(out_events, sched_out_state(prev_state), timestamp))
953 		return -1;
954 
955 	in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid);
956 	if (!in_events) {
957 		if (thread_atoms_insert(sched, sched_in))
958 			return -1;
959 		in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid);
960 		if (!in_events) {
961 			pr_err("in-event: Internal tree error");
962 			return -1;
963 		}
964 		/*
965 		 * Take came in we have not heard about yet,
966 		 * add in an initial atom in runnable state:
967 		 */
968 		if (add_sched_out_event(in_events, 'R', timestamp))
969 			return -1;
970 	}
971 	add_sched_in_event(in_events, timestamp);
972 
973 	return 0;
974 }
975 
976 static int latency_runtime_event(struct perf_sched *sched,
977 				 struct perf_evsel *evsel,
978 				 struct perf_sample *sample,
979 				 struct machine *machine)
980 {
981 	const u32 pid	   = perf_evsel__intval(evsel, sample, "pid");
982 	const u64 runtime  = perf_evsel__intval(evsel, sample, "runtime");
983 	struct thread *thread = machine__findnew_thread(machine, 0, pid);
984 	struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
985 	u64 timestamp = sample->time;
986 	int cpu = sample->cpu;
987 
988 	BUG_ON(cpu >= MAX_CPUS || cpu < 0);
989 	if (!atoms) {
990 		if (thread_atoms_insert(sched, thread))
991 			return -1;
992 		atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
993 		if (!atoms) {
994 			pr_err("in-event: Internal tree error");
995 			return -1;
996 		}
997 		if (add_sched_out_event(atoms, 'R', timestamp))
998 			return -1;
999 	}
1000 
1001 	add_runtime_event(atoms, runtime, timestamp);
1002 	return 0;
1003 }
1004 
1005 static int latency_wakeup_event(struct perf_sched *sched,
1006 				struct perf_evsel *evsel,
1007 				struct perf_sample *sample,
1008 				struct machine *machine)
1009 {
1010 	const u32 pid	  = perf_evsel__intval(evsel, sample, "pid"),
1011 		  success = perf_evsel__intval(evsel, sample, "success");
1012 	struct work_atoms *atoms;
1013 	struct work_atom *atom;
1014 	struct thread *wakee;
1015 	u64 timestamp = sample->time;
1016 
1017 	/* Note for later, it may be interesting to observe the failing cases */
1018 	if (!success)
1019 		return 0;
1020 
1021 	wakee = machine__findnew_thread(machine, 0, pid);
1022 	atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid);
1023 	if (!atoms) {
1024 		if (thread_atoms_insert(sched, wakee))
1025 			return -1;
1026 		atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid);
1027 		if (!atoms) {
1028 			pr_err("wakeup-event: Internal tree error");
1029 			return -1;
1030 		}
1031 		if (add_sched_out_event(atoms, 'S', timestamp))
1032 			return -1;
1033 	}
1034 
1035 	BUG_ON(list_empty(&atoms->work_list));
1036 
1037 	atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1038 
1039 	/*
1040 	 * You WILL be missing events if you've recorded only
1041 	 * one CPU, or are only looking at only one, so don't
1042 	 * make useless noise.
1043 	 */
1044 	if (sched->profile_cpu == -1 && atom->state != THREAD_SLEEPING)
1045 		sched->nr_state_machine_bugs++;
1046 
1047 	sched->nr_timestamps++;
1048 	if (atom->sched_out_time > timestamp) {
1049 		sched->nr_unordered_timestamps++;
1050 		return 0;
1051 	}
1052 
1053 	atom->state = THREAD_WAIT_CPU;
1054 	atom->wake_up_time = timestamp;
1055 	return 0;
1056 }
1057 
1058 static int latency_migrate_task_event(struct perf_sched *sched,
1059 				      struct perf_evsel *evsel,
1060 				      struct perf_sample *sample,
1061 				      struct machine *machine)
1062 {
1063 	const u32 pid = perf_evsel__intval(evsel, sample, "pid");
1064 	u64 timestamp = sample->time;
1065 	struct work_atoms *atoms;
1066 	struct work_atom *atom;
1067 	struct thread *migrant;
1068 
1069 	/*
1070 	 * Only need to worry about migration when profiling one CPU.
1071 	 */
1072 	if (sched->profile_cpu == -1)
1073 		return 0;
1074 
1075 	migrant = machine__findnew_thread(machine, 0, pid);
1076 	atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid);
1077 	if (!atoms) {
1078 		if (thread_atoms_insert(sched, migrant))
1079 			return -1;
1080 		register_pid(sched, migrant->tid, thread__comm_str(migrant));
1081 		atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid);
1082 		if (!atoms) {
1083 			pr_err("migration-event: Internal tree error");
1084 			return -1;
1085 		}
1086 		if (add_sched_out_event(atoms, 'R', timestamp))
1087 			return -1;
1088 	}
1089 
1090 	BUG_ON(list_empty(&atoms->work_list));
1091 
1092 	atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1093 	atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp;
1094 
1095 	sched->nr_timestamps++;
1096 
1097 	if (atom->sched_out_time > timestamp)
1098 		sched->nr_unordered_timestamps++;
1099 
1100 	return 0;
1101 }
1102 
1103 static void output_lat_thread(struct perf_sched *sched, struct work_atoms *work_list)
1104 {
1105 	int i;
1106 	int ret;
1107 	u64 avg;
1108 
1109 	if (!work_list->nb_atoms)
1110 		return;
1111 	/*
1112 	 * Ignore idle threads:
1113 	 */
1114 	if (!strcmp(thread__comm_str(work_list->thread), "swapper"))
1115 		return;
1116 
1117 	sched->all_runtime += work_list->total_runtime;
1118 	sched->all_count   += work_list->nb_atoms;
1119 
1120 	ret = printf("  %s:%d ", thread__comm_str(work_list->thread), work_list->thread->tid);
1121 
1122 	for (i = 0; i < 24 - ret; i++)
1123 		printf(" ");
1124 
1125 	avg = work_list->total_lat / work_list->nb_atoms;
1126 
1127 	printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %9.6f s\n",
1128 	      (double)work_list->total_runtime / 1e6,
1129 		 work_list->nb_atoms, (double)avg / 1e6,
1130 		 (double)work_list->max_lat / 1e6,
1131 		 (double)work_list->max_lat_at / 1e9);
1132 }
1133 
1134 static int pid_cmp(struct work_atoms *l, struct work_atoms *r)
1135 {
1136 	if (l->thread->tid < r->thread->tid)
1137 		return -1;
1138 	if (l->thread->tid > r->thread->tid)
1139 		return 1;
1140 
1141 	return 0;
1142 }
1143 
1144 static int avg_cmp(struct work_atoms *l, struct work_atoms *r)
1145 {
1146 	u64 avgl, avgr;
1147 
1148 	if (!l->nb_atoms)
1149 		return -1;
1150 
1151 	if (!r->nb_atoms)
1152 		return 1;
1153 
1154 	avgl = l->total_lat / l->nb_atoms;
1155 	avgr = r->total_lat / r->nb_atoms;
1156 
1157 	if (avgl < avgr)
1158 		return -1;
1159 	if (avgl > avgr)
1160 		return 1;
1161 
1162 	return 0;
1163 }
1164 
1165 static int max_cmp(struct work_atoms *l, struct work_atoms *r)
1166 {
1167 	if (l->max_lat < r->max_lat)
1168 		return -1;
1169 	if (l->max_lat > r->max_lat)
1170 		return 1;
1171 
1172 	return 0;
1173 }
1174 
1175 static int switch_cmp(struct work_atoms *l, struct work_atoms *r)
1176 {
1177 	if (l->nb_atoms < r->nb_atoms)
1178 		return -1;
1179 	if (l->nb_atoms > r->nb_atoms)
1180 		return 1;
1181 
1182 	return 0;
1183 }
1184 
1185 static int runtime_cmp(struct work_atoms *l, struct work_atoms *r)
1186 {
1187 	if (l->total_runtime < r->total_runtime)
1188 		return -1;
1189 	if (l->total_runtime > r->total_runtime)
1190 		return 1;
1191 
1192 	return 0;
1193 }
1194 
1195 static int sort_dimension__add(const char *tok, struct list_head *list)
1196 {
1197 	size_t i;
1198 	static struct sort_dimension avg_sort_dimension = {
1199 		.name = "avg",
1200 		.cmp  = avg_cmp,
1201 	};
1202 	static struct sort_dimension max_sort_dimension = {
1203 		.name = "max",
1204 		.cmp  = max_cmp,
1205 	};
1206 	static struct sort_dimension pid_sort_dimension = {
1207 		.name = "pid",
1208 		.cmp  = pid_cmp,
1209 	};
1210 	static struct sort_dimension runtime_sort_dimension = {
1211 		.name = "runtime",
1212 		.cmp  = runtime_cmp,
1213 	};
1214 	static struct sort_dimension switch_sort_dimension = {
1215 		.name = "switch",
1216 		.cmp  = switch_cmp,
1217 	};
1218 	struct sort_dimension *available_sorts[] = {
1219 		&pid_sort_dimension,
1220 		&avg_sort_dimension,
1221 		&max_sort_dimension,
1222 		&switch_sort_dimension,
1223 		&runtime_sort_dimension,
1224 	};
1225 
1226 	for (i = 0; i < ARRAY_SIZE(available_sorts); i++) {
1227 		if (!strcmp(available_sorts[i]->name, tok)) {
1228 			list_add_tail(&available_sorts[i]->list, list);
1229 
1230 			return 0;
1231 		}
1232 	}
1233 
1234 	return -1;
1235 }
1236 
1237 static void perf_sched__sort_lat(struct perf_sched *sched)
1238 {
1239 	struct rb_node *node;
1240 
1241 	for (;;) {
1242 		struct work_atoms *data;
1243 		node = rb_first(&sched->atom_root);
1244 		if (!node)
1245 			break;
1246 
1247 		rb_erase(node, &sched->atom_root);
1248 		data = rb_entry(node, struct work_atoms, node);
1249 		__thread_latency_insert(&sched->sorted_atom_root, data, &sched->sort_list);
1250 	}
1251 }
1252 
1253 static int process_sched_wakeup_event(struct perf_tool *tool,
1254 				      struct perf_evsel *evsel,
1255 				      struct perf_sample *sample,
1256 				      struct machine *machine)
1257 {
1258 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
1259 
1260 	if (sched->tp_handler->wakeup_event)
1261 		return sched->tp_handler->wakeup_event(sched, evsel, sample, machine);
1262 
1263 	return 0;
1264 }
1265 
1266 static int map_switch_event(struct perf_sched *sched, struct perf_evsel *evsel,
1267 			    struct perf_sample *sample, struct machine *machine)
1268 {
1269 	const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
1270 		  next_pid = perf_evsel__intval(evsel, sample, "next_pid");
1271 	struct thread *sched_out __maybe_unused, *sched_in;
1272 	int new_shortname;
1273 	u64 timestamp0, timestamp = sample->time;
1274 	s64 delta;
1275 	int cpu, this_cpu = sample->cpu;
1276 
1277 	BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0);
1278 
1279 	if (this_cpu > sched->max_cpu)
1280 		sched->max_cpu = this_cpu;
1281 
1282 	timestamp0 = sched->cpu_last_switched[this_cpu];
1283 	sched->cpu_last_switched[this_cpu] = timestamp;
1284 	if (timestamp0)
1285 		delta = timestamp - timestamp0;
1286 	else
1287 		delta = 0;
1288 
1289 	if (delta < 0) {
1290 		pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
1291 		return -1;
1292 	}
1293 
1294 	sched_out = machine__findnew_thread(machine, 0, prev_pid);
1295 	sched_in = machine__findnew_thread(machine, 0, next_pid);
1296 
1297 	sched->curr_thread[this_cpu] = sched_in;
1298 
1299 	printf("  ");
1300 
1301 	new_shortname = 0;
1302 	if (!sched_in->shortname[0]) {
1303 		sched_in->shortname[0] = sched->next_shortname1;
1304 		sched_in->shortname[1] = sched->next_shortname2;
1305 
1306 		if (sched->next_shortname1 < 'Z') {
1307 			sched->next_shortname1++;
1308 		} else {
1309 			sched->next_shortname1='A';
1310 			if (sched->next_shortname2 < '9') {
1311 				sched->next_shortname2++;
1312 			} else {
1313 				sched->next_shortname2='0';
1314 			}
1315 		}
1316 		new_shortname = 1;
1317 	}
1318 
1319 	for (cpu = 0; cpu <= sched->max_cpu; cpu++) {
1320 		if (cpu != this_cpu)
1321 			printf(" ");
1322 		else
1323 			printf("*");
1324 
1325 		if (sched->curr_thread[cpu]) {
1326 			if (sched->curr_thread[cpu]->tid)
1327 				printf("%2s ", sched->curr_thread[cpu]->shortname);
1328 			else
1329 				printf(".  ");
1330 		} else
1331 			printf("   ");
1332 	}
1333 
1334 	printf("  %12.6f secs ", (double)timestamp/1e9);
1335 	if (new_shortname) {
1336 		printf("%s => %s:%d\n",
1337 		       sched_in->shortname, thread__comm_str(sched_in), sched_in->tid);
1338 	} else {
1339 		printf("\n");
1340 	}
1341 
1342 	return 0;
1343 }
1344 
1345 static int process_sched_switch_event(struct perf_tool *tool,
1346 				      struct perf_evsel *evsel,
1347 				      struct perf_sample *sample,
1348 				      struct machine *machine)
1349 {
1350 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
1351 	int this_cpu = sample->cpu, err = 0;
1352 	u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
1353 	    next_pid = perf_evsel__intval(evsel, sample, "next_pid");
1354 
1355 	if (sched->curr_pid[this_cpu] != (u32)-1) {
1356 		/*
1357 		 * Are we trying to switch away a PID that is
1358 		 * not current?
1359 		 */
1360 		if (sched->curr_pid[this_cpu] != prev_pid)
1361 			sched->nr_context_switch_bugs++;
1362 	}
1363 
1364 	if (sched->tp_handler->switch_event)
1365 		err = sched->tp_handler->switch_event(sched, evsel, sample, machine);
1366 
1367 	sched->curr_pid[this_cpu] = next_pid;
1368 	return err;
1369 }
1370 
1371 static int process_sched_runtime_event(struct perf_tool *tool,
1372 				       struct perf_evsel *evsel,
1373 				       struct perf_sample *sample,
1374 				       struct machine *machine)
1375 {
1376 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
1377 
1378 	if (sched->tp_handler->runtime_event)
1379 		return sched->tp_handler->runtime_event(sched, evsel, sample, machine);
1380 
1381 	return 0;
1382 }
1383 
1384 static int perf_sched__process_fork_event(struct perf_tool *tool,
1385 					  union perf_event *event,
1386 					  struct perf_sample *sample,
1387 					  struct machine *machine)
1388 {
1389 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
1390 
1391 	/* run the fork event through the perf machineruy */
1392 	perf_event__process_fork(tool, event, sample, machine);
1393 
1394 	/* and then run additional processing needed for this command */
1395 	if (sched->tp_handler->fork_event)
1396 		return sched->tp_handler->fork_event(sched, event, machine);
1397 
1398 	return 0;
1399 }
1400 
1401 static int process_sched_migrate_task_event(struct perf_tool *tool,
1402 					    struct perf_evsel *evsel,
1403 					    struct perf_sample *sample,
1404 					    struct machine *machine)
1405 {
1406 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
1407 
1408 	if (sched->tp_handler->migrate_task_event)
1409 		return sched->tp_handler->migrate_task_event(sched, evsel, sample, machine);
1410 
1411 	return 0;
1412 }
1413 
1414 typedef int (*tracepoint_handler)(struct perf_tool *tool,
1415 				  struct perf_evsel *evsel,
1416 				  struct perf_sample *sample,
1417 				  struct machine *machine);
1418 
1419 static int perf_sched__process_tracepoint_sample(struct perf_tool *tool __maybe_unused,
1420 						 union perf_event *event __maybe_unused,
1421 						 struct perf_sample *sample,
1422 						 struct perf_evsel *evsel,
1423 						 struct machine *machine)
1424 {
1425 	int err = 0;
1426 
1427 	evsel->hists.stats.total_period += sample->period;
1428 	hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
1429 
1430 	if (evsel->handler != NULL) {
1431 		tracepoint_handler f = evsel->handler;
1432 		err = f(tool, evsel, sample, machine);
1433 	}
1434 
1435 	return err;
1436 }
1437 
1438 static int perf_sched__read_events(struct perf_sched *sched,
1439 				   struct perf_session **psession)
1440 {
1441 	const struct perf_evsel_str_handler handlers[] = {
1442 		{ "sched:sched_switch",	      process_sched_switch_event, },
1443 		{ "sched:sched_stat_runtime", process_sched_runtime_event, },
1444 		{ "sched:sched_wakeup",	      process_sched_wakeup_event, },
1445 		{ "sched:sched_wakeup_new",   process_sched_wakeup_event, },
1446 		{ "sched:sched_migrate_task", process_sched_migrate_task_event, },
1447 	};
1448 	struct perf_session *session;
1449 	struct perf_data_file file = {
1450 		.path = input_name,
1451 		.mode = PERF_DATA_MODE_READ,
1452 	};
1453 
1454 	session = perf_session__new(&file, false, &sched->tool);
1455 	if (session == NULL) {
1456 		pr_debug("No Memory for session\n");
1457 		return -1;
1458 	}
1459 
1460 	if (perf_session__set_tracepoints_handlers(session, handlers))
1461 		goto out_delete;
1462 
1463 	if (perf_session__has_traces(session, "record -R")) {
1464 		int err = perf_session__process_events(session, &sched->tool);
1465 		if (err) {
1466 			pr_err("Failed to process events, error %d", err);
1467 			goto out_delete;
1468 		}
1469 
1470 		sched->nr_events      = session->stats.nr_events[0];
1471 		sched->nr_lost_events = session->stats.total_lost;
1472 		sched->nr_lost_chunks = session->stats.nr_events[PERF_RECORD_LOST];
1473 	}
1474 
1475 	if (psession)
1476 		*psession = session;
1477 	else
1478 		perf_session__delete(session);
1479 
1480 	return 0;
1481 
1482 out_delete:
1483 	perf_session__delete(session);
1484 	return -1;
1485 }
1486 
1487 static void print_bad_events(struct perf_sched *sched)
1488 {
1489 	if (sched->nr_unordered_timestamps && sched->nr_timestamps) {
1490 		printf("  INFO: %.3f%% unordered timestamps (%ld out of %ld)\n",
1491 			(double)sched->nr_unordered_timestamps/(double)sched->nr_timestamps*100.0,
1492 			sched->nr_unordered_timestamps, sched->nr_timestamps);
1493 	}
1494 	if (sched->nr_lost_events && sched->nr_events) {
1495 		printf("  INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n",
1496 			(double)sched->nr_lost_events/(double)sched->nr_events * 100.0,
1497 			sched->nr_lost_events, sched->nr_events, sched->nr_lost_chunks);
1498 	}
1499 	if (sched->nr_state_machine_bugs && sched->nr_timestamps) {
1500 		printf("  INFO: %.3f%% state machine bugs (%ld out of %ld)",
1501 			(double)sched->nr_state_machine_bugs/(double)sched->nr_timestamps*100.0,
1502 			sched->nr_state_machine_bugs, sched->nr_timestamps);
1503 		if (sched->nr_lost_events)
1504 			printf(" (due to lost events?)");
1505 		printf("\n");
1506 	}
1507 	if (sched->nr_context_switch_bugs && sched->nr_timestamps) {
1508 		printf("  INFO: %.3f%% context switch bugs (%ld out of %ld)",
1509 			(double)sched->nr_context_switch_bugs/(double)sched->nr_timestamps*100.0,
1510 			sched->nr_context_switch_bugs, sched->nr_timestamps);
1511 		if (sched->nr_lost_events)
1512 			printf(" (due to lost events?)");
1513 		printf("\n");
1514 	}
1515 }
1516 
1517 static int perf_sched__lat(struct perf_sched *sched)
1518 {
1519 	struct rb_node *next;
1520 	struct perf_session *session;
1521 
1522 	setup_pager();
1523 
1524 	/* save session -- references to threads are held in work_list */
1525 	if (perf_sched__read_events(sched, &session))
1526 		return -1;
1527 
1528 	perf_sched__sort_lat(sched);
1529 
1530 	printf("\n ---------------------------------------------------------------------------------------------------------------\n");
1531 	printf("  Task                  |   Runtime ms  | Switches | Average delay ms | Maximum delay ms | Maximum delay at     |\n");
1532 	printf(" ---------------------------------------------------------------------------------------------------------------\n");
1533 
1534 	next = rb_first(&sched->sorted_atom_root);
1535 
1536 	while (next) {
1537 		struct work_atoms *work_list;
1538 
1539 		work_list = rb_entry(next, struct work_atoms, node);
1540 		output_lat_thread(sched, work_list);
1541 		next = rb_next(next);
1542 	}
1543 
1544 	printf(" -----------------------------------------------------------------------------------------\n");
1545 	printf("  TOTAL:                |%11.3f ms |%9" PRIu64 " |\n",
1546 		(double)sched->all_runtime / 1e6, sched->all_count);
1547 
1548 	printf(" ---------------------------------------------------\n");
1549 
1550 	print_bad_events(sched);
1551 	printf("\n");
1552 
1553 	perf_session__delete(session);
1554 	return 0;
1555 }
1556 
1557 static int perf_sched__map(struct perf_sched *sched)
1558 {
1559 	sched->max_cpu = sysconf(_SC_NPROCESSORS_CONF);
1560 
1561 	setup_pager();
1562 	if (perf_sched__read_events(sched, NULL))
1563 		return -1;
1564 	print_bad_events(sched);
1565 	return 0;
1566 }
1567 
1568 static int perf_sched__replay(struct perf_sched *sched)
1569 {
1570 	unsigned long i;
1571 
1572 	calibrate_run_measurement_overhead(sched);
1573 	calibrate_sleep_measurement_overhead(sched);
1574 
1575 	test_calibrations(sched);
1576 
1577 	if (perf_sched__read_events(sched, NULL))
1578 		return -1;
1579 
1580 	printf("nr_run_events:        %ld\n", sched->nr_run_events);
1581 	printf("nr_sleep_events:      %ld\n", sched->nr_sleep_events);
1582 	printf("nr_wakeup_events:     %ld\n", sched->nr_wakeup_events);
1583 
1584 	if (sched->targetless_wakeups)
1585 		printf("target-less wakeups:  %ld\n", sched->targetless_wakeups);
1586 	if (sched->multitarget_wakeups)
1587 		printf("multi-target wakeups: %ld\n", sched->multitarget_wakeups);
1588 	if (sched->nr_run_events_optimized)
1589 		printf("run atoms optimized: %ld\n",
1590 			sched->nr_run_events_optimized);
1591 
1592 	print_task_traces(sched);
1593 	add_cross_task_wakeups(sched);
1594 
1595 	create_tasks(sched);
1596 	printf("------------------------------------------------------------\n");
1597 	for (i = 0; i < sched->replay_repeat; i++)
1598 		run_one_test(sched);
1599 
1600 	return 0;
1601 }
1602 
1603 static void setup_sorting(struct perf_sched *sched, const struct option *options,
1604 			  const char * const usage_msg[])
1605 {
1606 	char *tmp, *tok, *str = strdup(sched->sort_order);
1607 
1608 	for (tok = strtok_r(str, ", ", &tmp);
1609 			tok; tok = strtok_r(NULL, ", ", &tmp)) {
1610 		if (sort_dimension__add(tok, &sched->sort_list) < 0) {
1611 			error("Unknown --sort key: `%s'", tok);
1612 			usage_with_options(usage_msg, options);
1613 		}
1614 	}
1615 
1616 	free(str);
1617 
1618 	sort_dimension__add("pid", &sched->cmp_pid);
1619 }
1620 
1621 static int __cmd_record(int argc, const char **argv)
1622 {
1623 	unsigned int rec_argc, i, j;
1624 	const char **rec_argv;
1625 	const char * const record_args[] = {
1626 		"record",
1627 		"-a",
1628 		"-R",
1629 		"-m", "1024",
1630 		"-c", "1",
1631 		"-e", "sched:sched_switch",
1632 		"-e", "sched:sched_stat_wait",
1633 		"-e", "sched:sched_stat_sleep",
1634 		"-e", "sched:sched_stat_iowait",
1635 		"-e", "sched:sched_stat_runtime",
1636 		"-e", "sched:sched_process_fork",
1637 		"-e", "sched:sched_wakeup",
1638 		"-e", "sched:sched_migrate_task",
1639 	};
1640 
1641 	rec_argc = ARRAY_SIZE(record_args) + argc - 1;
1642 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
1643 
1644 	if (rec_argv == NULL)
1645 		return -ENOMEM;
1646 
1647 	for (i = 0; i < ARRAY_SIZE(record_args); i++)
1648 		rec_argv[i] = strdup(record_args[i]);
1649 
1650 	for (j = 1; j < (unsigned int)argc; j++, i++)
1651 		rec_argv[i] = argv[j];
1652 
1653 	BUG_ON(i != rec_argc);
1654 
1655 	return cmd_record(i, rec_argv, NULL);
1656 }
1657 
1658 int cmd_sched(int argc, const char **argv, const char *prefix __maybe_unused)
1659 {
1660 	const char default_sort_order[] = "avg, max, switch, runtime";
1661 	struct perf_sched sched = {
1662 		.tool = {
1663 			.sample		 = perf_sched__process_tracepoint_sample,
1664 			.comm		 = perf_event__process_comm,
1665 			.lost		 = perf_event__process_lost,
1666 			.fork		 = perf_sched__process_fork_event,
1667 			.ordered_samples = true,
1668 		},
1669 		.cmp_pid	      = LIST_HEAD_INIT(sched.cmp_pid),
1670 		.sort_list	      = LIST_HEAD_INIT(sched.sort_list),
1671 		.start_work_mutex     = PTHREAD_MUTEX_INITIALIZER,
1672 		.work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER,
1673 		.sort_order	      = default_sort_order,
1674 		.replay_repeat	      = 10,
1675 		.profile_cpu	      = -1,
1676 		.next_shortname1      = 'A',
1677 		.next_shortname2      = '0',
1678 	};
1679 	const struct option latency_options[] = {
1680 	OPT_STRING('s', "sort", &sched.sort_order, "key[,key2...]",
1681 		   "sort by key(s): runtime, switch, avg, max"),
1682 	OPT_INCR('v', "verbose", &verbose,
1683 		    "be more verbose (show symbol address, etc)"),
1684 	OPT_INTEGER('C', "CPU", &sched.profile_cpu,
1685 		    "CPU to profile on"),
1686 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1687 		    "dump raw trace in ASCII"),
1688 	OPT_END()
1689 	};
1690 	const struct option replay_options[] = {
1691 	OPT_UINTEGER('r', "repeat", &sched.replay_repeat,
1692 		     "repeat the workload replay N times (-1: infinite)"),
1693 	OPT_INCR('v', "verbose", &verbose,
1694 		    "be more verbose (show symbol address, etc)"),
1695 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1696 		    "dump raw trace in ASCII"),
1697 	OPT_END()
1698 	};
1699 	const struct option sched_options[] = {
1700 	OPT_STRING('i', "input", &input_name, "file",
1701 		    "input file name"),
1702 	OPT_INCR('v', "verbose", &verbose,
1703 		    "be more verbose (show symbol address, etc)"),
1704 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1705 		    "dump raw trace in ASCII"),
1706 	OPT_END()
1707 	};
1708 	const char * const latency_usage[] = {
1709 		"perf sched latency [<options>]",
1710 		NULL
1711 	};
1712 	const char * const replay_usage[] = {
1713 		"perf sched replay [<options>]",
1714 		NULL
1715 	};
1716 	const char * const sched_usage[] = {
1717 		"perf sched [<options>] {record|latency|map|replay|script}",
1718 		NULL
1719 	};
1720 	struct trace_sched_handler lat_ops  = {
1721 		.wakeup_event	    = latency_wakeup_event,
1722 		.switch_event	    = latency_switch_event,
1723 		.runtime_event	    = latency_runtime_event,
1724 		.migrate_task_event = latency_migrate_task_event,
1725 	};
1726 	struct trace_sched_handler map_ops  = {
1727 		.switch_event	    = map_switch_event,
1728 	};
1729 	struct trace_sched_handler replay_ops  = {
1730 		.wakeup_event	    = replay_wakeup_event,
1731 		.switch_event	    = replay_switch_event,
1732 		.fork_event	    = replay_fork_event,
1733 	};
1734 	unsigned int i;
1735 
1736 	for (i = 0; i < ARRAY_SIZE(sched.curr_pid); i++)
1737 		sched.curr_pid[i] = -1;
1738 
1739 	argc = parse_options(argc, argv, sched_options, sched_usage,
1740 			     PARSE_OPT_STOP_AT_NON_OPTION);
1741 	if (!argc)
1742 		usage_with_options(sched_usage, sched_options);
1743 
1744 	/*
1745 	 * Aliased to 'perf script' for now:
1746 	 */
1747 	if (!strcmp(argv[0], "script"))
1748 		return cmd_script(argc, argv, prefix);
1749 
1750 	symbol__init();
1751 	if (!strncmp(argv[0], "rec", 3)) {
1752 		return __cmd_record(argc, argv);
1753 	} else if (!strncmp(argv[0], "lat", 3)) {
1754 		sched.tp_handler = &lat_ops;
1755 		if (argc > 1) {
1756 			argc = parse_options(argc, argv, latency_options, latency_usage, 0);
1757 			if (argc)
1758 				usage_with_options(latency_usage, latency_options);
1759 		}
1760 		setup_sorting(&sched, latency_options, latency_usage);
1761 		return perf_sched__lat(&sched);
1762 	} else if (!strcmp(argv[0], "map")) {
1763 		sched.tp_handler = &map_ops;
1764 		setup_sorting(&sched, latency_options, latency_usage);
1765 		return perf_sched__map(&sched);
1766 	} else if (!strncmp(argv[0], "rep", 3)) {
1767 		sched.tp_handler = &replay_ops;
1768 		if (argc) {
1769 			argc = parse_options(argc, argv, replay_options, replay_usage, 0);
1770 			if (argc)
1771 				usage_with_options(replay_usage, replay_options);
1772 		}
1773 		return perf_sched__replay(&sched);
1774 	} else {
1775 		usage_with_options(sched_usage, sched_options);
1776 	}
1777 
1778 	return 0;
1779 }
1780